[Intel-gfx] [PATCH i-g-t 2/5] gem_wsim: Round mmap to page size

Tvrtko Ursulin tvrtko.ursulin at linux.intel.com
Thu Feb 28 14:41:23 UTC 2019


On 28/02/2019 14:25, Chris Wilson wrote:
> Quoting Tvrtko Ursulin (2019-02-28 14:18:25)
>> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>
>> mmap(2) mandates size is page aligned.
>>
>> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>> ---
>>   benchmarks/gem_wsim.c | 7 ++++---
>>   1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/benchmarks/gem_wsim.c b/benchmarks/gem_wsim.c
>> index 0a5abc08d8c2..57ceb983cf82 100644
>> --- a/benchmarks/gem_wsim.c
>> +++ b/benchmarks/gem_wsim.c
>> @@ -737,6 +737,7 @@ init_bb(struct w_step *w, unsigned int flags)
>>   {
>>          const unsigned int arb_period =
>>                          get_bb_sz(w->preempt_us) / sizeof(uint32_t);
>> +       const unsigned int mmap_len = ALIGN(w->bb_sz, 4096);
>>          unsigned int i;
>>          uint32_t *ptr;
> 
> You only need to do it for
> 
> ww->bb_sz = ALIGN(get_bb_sz(), 4096);
> 
> All batch lengths are start = ww->bb_sz - get_bb_sz() and so remain
> correct. Right?

I think so. I have one more assignment site of w->bb_sz in the upcoming 
code so it was 2 : 2 and I flipped a coin. Actually I wanted to make it 
explicit what I am fixing - the mmap size. But now you'll say batch size 
is also implicitly rounded.. oh well.. I prefer the explicit round up at 
mmap time.

Regards,

Tvrtko


More information about the Intel-gfx mailing list